A multimedia database supports English distance learning

Information Sciences 158 (2004) 189–208 www.elsevier.com/locate/ins

A multimedia database supports English distance learning Ying-Hong Wang *, Chih-Hao Lin Department of Computer Science and Information Engineering, Graduate Institute of Information Engineering, Tamkang University, Tamsui, Taipei 251, Taiwan Received 20 February 2003; received in revised form 27 June 2003; accepted 15 July 2003

Abstract This work presents a novel English distance learning system that was developed through multimedia database and Internet technologies called English multimedia corpus. The system includes English articles, dialogs, and videos. A student can study English writing and reading as well as view Web browser listings to connect the Corpus server. In the system, semantic query and ‘‘link grammar’’ are applied to construct the English multimedia corpus system. Furthermore, it promotes the query level from keyword-base and content-based query to a semantic level. The main function of this system is to query the English sentence pattern through keywords from the English multimedia corpus. The other function is to detect grammatical errors in written English. Thus, the system not only teaches English grammar, but also, due to its database, allows teachers to understand the most frequent mistakes. Ó 2003 Elsevier Inc. All rights reserved. Keywords: Multimedia database; Link grammar; Distance learning; Linguistic corpus

1. Introduction Traditional databases always store the character and numerical data. A primary function manages this basic data, for example, Students, School data,

*

Corresponding author. E-mail addresses: [email protected] (Y.-H. Wang), [email protected] (C.-H. Lin). 0020-0255/$ - see front matter Ó 2003 Elsevier Inc. All rights reserved. doi:10.1016/j.ins.2003.07.002

190

Y.-H. Wang, C.-H. Lin / Information Sciences 158 (2004) 189–208

or Companies’ employee and financial data. Following computer-based technological advancements, multiple types of data, such as image, audio, video, and hypermedia documents, are applied to represent computer information. Thus, database technologies must support multimedia data, which is known as a multimedia database. It includes many facets, such as content-based retrieval [6,15], shape detection and object recognition [8,10,11,18]. Meanwhile, these technical aspects promote its numerous application domains. Many examples of distance learning, digital television, and distance medicine are implemented based on those techniques. This study presents an Internet-based English learning environment that applies multimedia database technologies. Learning English is always a substantial difficulty for Chinese students. Some experts recommend that the best way to learn English is to establish a good study environment and practice it through several various approaches. This should make it a satisfying learning experience. Based on this viewpoint, teachers should provide students with a better environment to learn English. Distance learning based on Internet is one of the best solutions to create an English learning environment. Currently, the highly distributed computing environment has been supported and applied popularly. Networking services should perform as tutoring tools. Numerous innovative approaches in this field have been investigated. Through implementing computer-mediated education, many advocates emphasize its positive aspects and English learning tutoring systems, which are computer-based, have been developed by numerous academic research groups [19–31]. Incidentally, students learning a foreign language through computing devices might encounter many difficulties. Some articles have discussed whether students’ frustrations inhibit their educational opportunity [20,21,23]. These frustrations were based on three interrelated sources: lack of prompt feedback, ambiguous Web instructions, and technical problems. Hence, an easy-to-use English learning system was developed. The proposed system contains several preparations, which are relevant to English learning. While treating natural language processing, the system is constructed based on link grammar. However, in order to separate it from conventional approaches, modifications were included to improve its ability. The proposed English-learning corpus stores mistakes that students make. That is, all errors that students will probably make on a particular question are determined in advance. Each type of mistake has correcting suggestions and error descriptions. Corpus cannot only provide suggestions but it also records the mistakes that Chinese students might make in English. Another function of this Internet-based multimedia corpus is that a user can locate sample sentences patterns from movie scripts, live dialogs and textbook reading. Each sentence from live dialog and movie captions is parsed and analyzed by link grammar, which includes special tags. The end product is then stored within


191

this proposed database system. Consequently, based on semantic retrieval, the system subscribes to query methodology. In this paper, English sentence construction will be analyzed first. Notably, the sentence pattern is divided into nine classes. Secondly, each sentence pattern is parsed by link grammar, in which every pattern has a particular tag set. Consequently, the system constructs the query engine with these essential tag sets. In general, similar learning systems store single sentences or vocabularies and facilitate query through keywords [19–22,24,25,27–31]. Semantic-based retrieval technologies are rarely used. This work discusses how to use the appended information, the special tag sets, which link grammar generated to advance the query function from keyword-based to semantic-based [1,2,7]. Several web sites regarding learning corpus [19–22], and study several papers in addition to those that pertain to link grammar [4,9,13,16] were surveyed. Furthermore, Schulenburg proposed realistic feedback processing [9], Brill developed machine learning and automatic linguistic analysis [12], and Chang presented automatic linguistic resolution: framework and applications [3]. Other similar systems are referenced [4,5,14,16,17]. In Section 2, related works regarding link grammar is discussed. The system analysis for the proposed learning functions is introduced in Section 3. Section 4 presents the architecture of the proposed system and evaluation with other systems; there are some functions are same to ours. Finally, Section 5 contains the conclusion and discussions for future research.

2. Theoretical backgrounds Link grammar is an English grammar parser system that was proposed by The School of Computer Science of Carnegie Mellon University (CMU). Link grammar is a context-free formula to describe natural language [32]. Link grammar consists of a set of words, which are the terminal symbols of the grammar, and each has a linking requirement. The linking requirements of each word are gathered in a dictionary. Fig. 1 illustrates the linking requirements through a simple dictionary for the words, a/the, cat/mouse, John, ran, chased. Each of the intricately shaped boxes is labeled connector. A pair of compatible connectors will join, and only one of those attached to a given black dot must be satisfied. Fig. 2 depicts how the linking requirements are satisfied in the sentence, ‘‘The cat chased a mouse’’. The linkage can be perceived as a graph and the words can be treated as vertices, which are connected by labeled arcs. Thus, the graph is connected and planar. The labeled arcs that connect word to others on either their left or right are links. A valid parse is called a linkage. Thus, the following is a simplified form of the diagram indicating that the cat chased a mouse is part of this language (Fig. 3).

192

Y.-H. Wang, C.-H. Lin / Information Sciences 158 (2004) 189–208 O D

D

S

O

S

cat mouse

a the

S

John

S

ran

O

chased

Fig. 1. Words and connectors in a dictionary.

O D

the

D

S

O D

S

cat

chased

D

a

mouse

Fig. 2. All linking requirements are satisfied to form a linkage.

O D

S D

the

cat

chased

a

mouse

Fig. 3. The simplified form of Fig. 2.

Table 1 presents an abridged dictionary, which encodes linking requirements of the above example. The linking requirement for each word is expressed as a formula involving that includes the operators &, and or, parentheses, and connector names. The + or ) suffix on a connector indicates the direction in which the matching connector must lie be placed. Thus, the farther left a connector is in the within an expression phrase, the nearer the word to which it connects must be. A sequence of words is a sentence defined by grammar, if links can be established among the words so as to satisfy the formula of each word. It includes the following meta-rules: Planarity: Links are drawn above the sentence and do not cross. Connectivity: Links connect all the words in the sequence. Ordering: When the connectors of a formula are traversed from left to right, the words to which they connect proceed from near to far. Namely, consider a word, and consider two links connecting that word to the


193

Table 1 The words and linking requirements in a dictionary Words

Formula

a/the cat/mouse John ran chased

D+ D) & (O) or S+) O) or S+ S) S) & O+

word on its left. The link connecting the closest word (the shorter link) must satisfy a connector that appears to the left (in the formula) of that connector in the other word. The same process occurs for a link to the right. Exclusion: No two links may connect the same pair of words. The use of a formula to specify a link grammar dictionary is convenient for creating natural language grammars, however it is cumbersome for mathematical analysis there of, and as well as in describing algorithms to parse link grammar. An alternate method of expressing link grammar is known as disjunctive form, in which each word has an associated set of disjuncts. A disjunct is denoted as ððL1 ; L2 ; . . . ; Lm ÞðRn ; Rn1 ; . . . ; R1 ÞÞ where L1 ; L2 ; . . . ; Lm and Rn ; Rn1 ; . . . ; R1 are the connectors that must connect to the left and right, respectively. Thus, simply rewriting each disjunct and combining them all with the or operator to create an appropriate formula translates a link grammar in from a disjunctive to a standard form can be completed by as follows: ðL1 &L2 & . . . &Lm &R1 &R2 & . . . &Rn Þ Enumerating all ways that the formula cab be satisfied translates a formula into a set of disjuncts. For example, the formula ðA or ð ÞÞ&D &ðB þ or ð ÞÞ&ðO or SþÞ corresponds to the following eight disjuncts, may be used in some linkages: ððA; DÞ ðS; BÞÞ ððA; D; OÞ ðBÞÞ ððA; DÞ ðSÞÞ ððA; D; OÞ ð ÞÞ ððDÞ

ðS; BÞÞ

ððD; OÞ ðBÞÞ ððDÞ ðSÞÞ ððD; OÞ

ð ÞÞ

194


3. System analysis The proposed system contains two varieties of learning functions, or two sub-systems. These sub-systems are English pattern query system and English sentence error-detect system. The detail will be discussed in the following. 3.1. English pattern query system To develop this system, English sentence patterns were analyzed. Based on general analysis, there are five English sentence patterns, as follows: (1) (2) (3) (4) (5)

Simple sentence pattern. Negative sentence pattern. Interrogative sentence pattern. WH question sentence pattern (What, When, Where,. . .). Imperative sentence pattern.

They can then be divided by tense, as follows: Simple pattern The simple present tense: He writes a letter everyday. The simple past tense: He wrote a letter yesterday. The simple future tense: He will write a letter tomorrow. Perfective pattern The present perfective tense: He has written a letter. The past perfective tense: He had written a letter when I arrived. The future perfective tense: He will have written a letter before I arrive. Continuous pattern The present continuous tense: He is writing a letter now. The past continuous tense: He was writing a letter when I arrived. The future continuous tense: He will be writing a letter when I arrive. Perfective proceed pattern The present perfective continuous tense: He is writing a letter now. The past perfective continuous tense: He was writing a letter when I arrived. Link grammar parses these sentences to determine the label correlation of their sentence patterns (Table 2). Subsequently, each pattern receives a label constituent type. This characteristic parses each sentence in the system, which stores the resulting label set. For example, (Ss, ppf, pg) is a label set that was achieved and stored following


195

Table 2 Simple sentence pattern and link grammar tags Active voice

Passive voice

Simple pattern

Present Past Future

Ss or Sp Ss or Sp Ss or Sp + I

Ss or Sp + Pv Ss or Sp + Pv Ss or Sp + Ix + Pv

Continuous pattern

Present Past Future

Ss or Sp + PP Ss or Sp + PP Ss + If + PP

Ss or Sp + ppf + Pv Ss or Sp + ppf + Pv Ss or Sp + If + ppf + Pv

Proceed pattern

Present Past Future

Ss or Sp + Pg Ss or Sp + Pg Ss or Sp + Ix + Pg

Ss or Sp + Pg + Pv Ss or Sp + Pg + Pv None

Perfective continuous pattern

Present Past

Ss or Sp + ppf + Pg Ss or Sp + ppf + Pg

Illustrate: Ss and Sp: connects subject nouns to finite verbs; I: connects infinitive verb forms to certain words such as modal verbs and ‘‘to’’; PP: connects forms of ‘‘have’’ with past participles; PV: connects forms of the verb‘‘be’’ to past participles; Pg: connects forms of the verb ‘‘be’’ to present participles; ppf: connects forms of the verb ‘‘be’’ to past participles ‘‘been’’.

link grammar parsing. Then, a user can employ query language to arrive at the English sentence pattern they desire. Initially, to ensure that a pattern is accurate, analysis of only one type of pattern is performed. The system can then be upgraded to handle multi-patterns. The system can also process multiple queries in several types of patterns. However, this has some difficulties, which are discussed in Section 3.2. 3.2. English sentence error detect system In this section, the Enhanced Link Grammar, which is based on a dictionary with modified disjuncts, is illustrated. Herein, word linkage and the applications of Enhanced Link Grammar are discussed, respectively. Modified Dictionary In general, the formula of each word is complicated. Initially, the word that requires connector changes is analyzed. Various formulas impose distinct constraints on the proposed system. Thus, although changes may improve the system, many may not. To illustrate, simple words are tested prior to further explanation. For example, some verbs must be followed by an infinitive –to, a gerund or a participle. However, several verbs can accept several varying types. For example, ‘‘We enjoy walking in the rain.’’ is accurate, but ‘‘We enjoy to walk in the rain.’’ is not. However, the link of the word ‘‘enjoy’’ can be altered to accept the infinitive –to.

196


Fig. 4. The original link grammar parser.

Fig. 5. The modified link grammar parser.

A portion of the formula is changed from ‘‘(Pg)’’ to ‘‘(Pg or TO e9+)’’. Hence, the word ‘‘enjoy’’ can be linked to not only a gerund but also an infinitive –to with the additional connector ‘‘TO e9+’’. Notably, this connector contains a subscript ‘‘ e9’’, which indicates there is a type 9 error. The type 9 error indicates the mistake is happened when a transitive verb is used as intransitive verb. Subsequently, to the link grammar with original dictionary (Fig. 4) and modified dictionary (Fig. 5), respectively, there are another two sentences ‘‘We enjoy swimming.’’ and ‘‘We enjoy to swim.’’


197

Fig. 4 presents that the link grammar that has the original dictionary cannot accept an inaccurate sentence. That is, an error message ‘‘No complete linkage found.’’ will appear. However, Fig. 5 presents the experimental result when the same two sentences are parsed by the link grammar that contains the modified dictionary. Thus, ‘‘enjoy’’ can link to either a gerund or an infinitive –to. Incidentally, there are 10 types of catalogued error tags for error detection. In Section 4, the error tag, including the links that can and cannot be modified, is discussed.

4. The system architecture and examples 4.1. System architecture Corpus is a multimedia database that provides a dictionary as well as record errors, comments, and system feedback. Similar to English essays, dialog, sentences, words, terms, and attributes, the proposed English-learning corpus also has positive study data. It promotes language teaching and research. In addition to the original data, it also stores multimedia data, including instructional films, and audio clips, movies, music videos. Therefore, it provides a vivid, attractive learning environment. This system, presented in Figs. 9 and 10, is a multimedia corpus that provides four main functions of basic language learning; they are listen, speech, read, and write. That is a user can read an essay, listen to Standard English pronunciation, write compositions, practice dialog and recite text. The database is a vital tool for linguistic researchers. Philologists can analysis mistakes that English as second language students typically make. This system can be used over the Internet to access the data from the multimedia learning corpus. Fig. 6 illustrates how a user can access this system from any location. As stated, the proposed system contains two sub-systems, which are English pattern query system and English sentence error detect system. Notably, they share the same database. Fig. 7 depicts the architecture of the ‘‘English pattern query system’’. Link grammar with the original dictionary parses the data, which produces link labels. The system then analyses and filters the link labels, which are stored into the database. When a user queries a sentence pattern, they use the interface of the system to obtain the desired pattern. Subsequently, the system will transform the query into readable data-type and, if possible, it will locate the relevant link label within the database. The system will retrieve the data and send it to the multimedia player on the Web. The advantage of this is that a user can replay messages several times until they clearly understand the correct

198


Fig. 6. System architecture overview.

English pattern Query Function

data input Data/Sentence

data transform

Link grammar Parser (Original)

DataBase

Label analysis & filter

multimedia player on the web

Output

Fig. 7. English pattern query system architecture.

pronunciation of the sentence, as well as the context in which the sentence should be used.


199

English sentence input

Link grammar parser (enhanced)

Learner Corpus Label analysis & filter

Is there error tag appear

Yes

access the data in the learner corpus

put the error sentence into the database

No

output

output DataBase

Fig. 8. English sentence error detect system architecture.

Fig. 8 depicts the architecture of the other second sub-system, ‘‘English sentence error detect system’’. The user can input a sentence into the system. Following analysis, the link grammar with the modified dictionary will parse the sentence, and filter the link labels if error tags are detected. The system will locate the description and error examples within the learner corpus and then display that information on the web page. However, if the system fails to understand an error, it may not have been defined or is too complex. Those sentences will be stored in the database for philologist analysis and further subsequent system enhancement. 4.2. System examples In this section, operation examples of the system, which is a web-based user interface, are presented. Fig. 9 depicts the interface of the English pattern query system. This provides users with an easy-to-use interface; that is they simply need to drop down the menu and select the pattern that they want to read. The user can input keywords to search the sentence, which they want to read, or drop down the menu to choose the sentence patterns or time terms. For example, user can search sentence that is in ‘‘The past continuous tense’’ with the keyword

200


Fig. 9. English pattern query system interface.

Fig. 10. Keyword pattern and Drop down menu.

‘‘writing’’ (Fig. 10). User can use the tools in the system interface. And then the system will then return the query result with the multimedia data on the web browser (Fig. 11). In the system, there are many kinds of multimedia data type in the corpus. When the sentence query by the user, the link of the multimedia data also can be show on the web (Fig. 12). If the user is interesting in some sentences or they


201

Fig. 11. Query result.

Fig. 12. Play multimedia data on the web.

want to watch, read or listen to them in multimedia representation, these data can be played on the web too. It is very convenient for the user that they can read sentence and listen the voice at the same time. And then Fig. 13 presents the interface of the English sentence error detects system.

202


Fig. 13. English sentence error detect system interface.

Fig. 14. Correct sentence result.

When the user makes a sentence, they can input to the interface of ‘‘English Grammar Error Detect System’’. After a sentence is input, link grammar with modified dictionary will determine errors. Fig. 14 illustrates an example of a correctly input English sentence, ‘‘the little boy always walks to school’’. The system can check the grammar of this sentence. If there is nothing wrong with sentence the system will return correct message on the web.


203

Fig. 15. Error sentence result.

However, when the user makes an incorrect sentence like ‘‘the little boy always walk to school’’. The system will find the sentence has something wrong between with these two words ‘‘boy’’ and ‘‘walk’’. So system will change these two words to red that can remind the user they make a mistake with these two words. Fig. 15 presents the system’s response when an error is made. In the back of the system, when the user input an incorrect sentence. The system also searches the Learner Corpus. And input this sentence into the Learner Corpus. This is very important information for the system or the linguist. This information is very useful for them that can promote the function of our system and can help linguist to find the answer why the student can make this kind of mistake. This system not only can detect the grammar error but also can suggest the user what mistakes they made and help them don’t do this next time. User pushes the ‘‘suggestion’’ button then the system can show the suggestion, which is like Fig. 16. In this example, system shows the incorrect sentence that is made by the user, and what kind of mistake and description of error type. They also show correct and incorrect sentences of examples. 4.3. Evaluation Several related systems are surveyed as follows: IwiLL [33], which is a web-based language-learning platform, consists of several tightly interwoven components each designed by the IWiLL team. This is a learner corpus for the students in Taiwan. Student can write an article on

204


Fig. 16. Error suggestion.

the web and then send to the teacher. The teacher can check also the homework on the web. The teacher not only can check spelling error and grammar mistake but also semantic misemployed. It also provides some multimedia data to help the student learn English. And store the article, which is made by the student into the learner corpus. However, it is not an automatic system. Teachers need to keep on-line frequently. American National Corpus [34–36], which includes, ideally, a balanced representation of texts and transcriptions of spoken data. And the samples of other major languages of North American, especially Spanish and French Canadian should also comprise a portion of the corpus and, ideally, be aligned to parallel translations in English. ANC comprise approximately 100 million words that include additional texts from a wide range of styles and domains. It is a word corpus only. It cannot record the learning problems of students. British National Corpus [37], which is a 100 million words collection of samples of written and spoken language from a wide range of sources, designed to represent a wide cross-section of current British English, both spoken and written. BNC is an online service that is difference form ANC. It is also a word corpus only. Its drawbacks are same as American National Corpus. CRITIQUE, this system is provide by Yael Ravin [38]. The method of this system are grammar error detect and style weaknesses. This system is linguistic base that is based on rule base. The system analysis the grammar and styles weaknesses, as terms error and weakness suggest. CRITIQUE can detect five categories grammar errors and eight categories style weaknesses. This system uses the rule base methodology. But it is only an application; user cannot use

Corpus

ELCS

IwiLL

ANC

BNC

CRITIQUE

MS Grammar checker

Lerner corpus

Lerner corpus

Standard corpus

Standard corpus

N

N

1000 millions

1000 millions

Words Grammar check Number disagreement Wrong pronoun Wrong verb from Wrong article Comparative degree Punctuation Time terms

Automatic Y Y Y Y Y Y Y

Manually

N – – – – – – –

N – – – – – – –

Automatic Y Y Y N N Y N

Automatic Y Y Y Y Y Y Y

Web application Support multimedia Semantic query

Y Y Y

Y Y N

N N N

Ya

N

Y N Semantic annotation –

N N N

Semantic parser

N N Semantic annotation –

N

N

: checked by teacher. a Keep on going system.


Table 3 Evaluation with other systems

205

206


this system on the web. The system can record the information from the user that cannot upgrade their functions. Microsoft Office Word grammar checker, this is a well know system. When you using the word system, there are an agent ‘‘paper clip’’ or something else to help you. If you make a spell error or grammar mistake, the agent will show the error and can correct the mistake. The system uses the several dictionaries like ActiveGrammarDictionary, ActiveHyphenationDictionary, ActiveSpellingDictionary, and ActiveThesaurusDictionary properties to return the corresponding Dictionary objects. And the system uses the statistically methodology to detect and correct the errors [39]. But this system cannot collect the user information when they use the system. That is difficult to upgrade the capability for next version. Table 3 shows the evaluations our system with other systems.

5. Conclusions and future work 5.1. The contribution and remarks After following analysis and experimentation, the system can easily use the label set, which is stored in the corpus and parsed with link grammar. Primarily, it assists junior or senior high school students to locate proper English sentence patterns. Numerous linguistic experts recommend that the best way to improve English speaking ability is to establish a realistic learning environment and practice English through several distinct approaches. The proposed interactive multimedia system promotes this type of learning. Sleator and Temperley developed Link Grammar, which is a word based parsing mechanism. Although the system was though to be an accurate grammar checker, unlike the proposed system, it fails to focus on fault tolerance ability, which is particularly useful for non-native English speaking students. The proposed system can parse sentences that the original link grammar cannot. Notably, our modified dictionary has progressed steadily. It appears that the idea proposed herein can be applied to other natural language processing and its corresponding applications. The system not only queries English sentence patterns, but also parses input sentences on-line. It is important to philologists to analysis sentences generated by students, so they can easily locate common or special mistakes. Subsequently, they can alter the focus of their course for a complete learning environment. This system provides a better and more interactive environment for both teachers and students. Furthermore, it combines the humanistic education with technology and advances these two sciences for superior achievement.


207

5.2. Future works Generally, words are a basic unit of communication that is common to all natural language texts and speech-processing activities. When teaching English, teachers always want to know the types of mistakes that students may make. Thus, according to various functional necessities, the proposed system can be extended to cover encompass more scalable ranges. The notions presented herein can aid in the development of other similar applications. XML has self-descriptive characteristics, which has resulted in increasing data interchange applications. These applications and research areas may become a popular subject in the future. We are undertaking a semantic Link Grammar parser now. This system can parse sentence use the same way like link grammar methodology. But the system can get semantic information from the sentence. That is very useful for next step to understand sentence, can translation to another language and other applications.

References [1] W. Al-Khatib, Y. Francis Day, A. Ghafoor, P. Bruce Berra, Semantic modeling and knowledge representation in multimedia databases, IEEE Trans. Knowledge Data Eng. 11 (1) (1999). [2] D. Beeferman, A. Berger, J. Lafferty, Text segmentation using exponential models, in: Proceedings of the Second Conference on Empirical Methods in NLP, Providence, RI, 1997. [3] E. Brill, Machine learning and automatic linguistic analysis: the next step, Acoust. Speech Signal Process., Proceedings of the 1998 IEEE International Conference, vol. 2, 1998, pp. 1033–1036. [4] C.M. Chung, Y.-K. Wang, Y.-H. Wang, C.-H. Lin, The multimeaia corpus based on link grammar, National Computer Symposium 99, vol. 1, pp. A-300–304. [5] C.M. Chung, Y.-K. Wang, Y.-H. Wang, C.-H. Lin, The design of multimedia database for english learning application, in: Proceeding of 2000 International Conference on Chinese Language Computing (ICCLC2000), pp. 77–81. [6] S.-F. Chang, W. Chen, H.J. Meng, H. Sundaram, D. Zhong, A fully automated content-based video search engine supporting spatiotemporal queries, IEEE Trans. Circ. Syst. Video Technol. 8 (1998) 602–615. [7] S.W.K. Chan, Automatic linguistic resolution: framework and applications, Syst. Man Cybernet., IEEE Internation Conference, vol. 1, 1996, pp. 625–630. [8] S.-K. Chang, A. Hsu, Image Information systems: where do we go from here?, IEEE Trans. Knowledge Data Eng. 4 (1992) 431–442. [9] D. Grinberg, J. Lafferty, D. Sleator, A robust parsing algorithm for link grammars, in: Proceedings of the Fourth International Workshop on Parsing Technologies, Prague and Karlovy Vary, September 1995, pp. 111–125. [10] F. Myron, S. Harpreet, N. Wayne, A. Jonathan, H. Qian, D. Byron, G. Monika, H. Jim, L. Denis, P. Dragutin, Query by Image and Video Content: The QBIC System, Computer, vol. 28 (9), IEEE, Los Alamitos CA, USA, September 1995, pp. 23–32.

208


[11] Y. Nakamura, T. Kanade, Semantic analysis for video contents extraction––spotting by association in new video, ACM Multimedia Conference, 1997. [12] D. Schulenburg, Sentence processing with realistic feedback, Neural Networks, IJCNN, International Joint Conference, vol. 4, 1992, pp. 661–666. [13] D. Sleator, D. Temperley, Parsing English with a link grammar, technical report CMU-CS-91196, Department of Computer Science, Carnegie Mellon University, 1991. [14] C. Yamada, H. Sumiyoshi, Y.-B. Kim, M. Shibata, TV program skimming method using similarity of scene descriptions, SPIE Conference on Multimedia Storage and Archiving Systems III, November 1998. [15] A. Yoshitaka, T. Ichikawa, A survey on content-based retrieval for multimedia databases, IEEE Trans. Knowledge Data Eng. 11 (1) (1999) 81–93. [16] Y.-K. Wang, Y.-H. Wang, H.-P. Lin, C.-F. Chao, An English-Learning Application based on extended link grammar formalism, National Computer Symposium 99, vol. 2, pp. B-306–310. [17] I.W. Agius, M.C. Angelides, Developing knowledge-based intelligent multimedia tutoring systems using semantic content-based modeling, Artif. Intell. Rev. 13 (1999) 55–83. [18] J.-K. Wu, Content-based indexing of multimedia databases, IEEE Trans. Knowledge Data Eng. 9 (6) (1997). [19] EFLWEB Home Page. Available from . [20] English grammar clinic. Available from . [21] Foreign Languages for Travelers. Available from . [22] G.R. Sampson: SUSANNE Scheme. Available from . [23] Interactive Javascript Quizzes for ESL Students. Available from . [24] Language Dictionaries and Translators. Available from . [25] Linguistic Data Consortium LDC. Available from . [26] Link Grammar. Available from . [27] Longman Dictionaries Home. Available from . [28] Merriam-Webster Online. Available from . [29] On line English Grammar. Available from . [30] Project Gutenberg Official Home Site. Available from . [31] YourDictionary.COM. Available from . [32] J. Lafferty, D. Sleator, D. Temperley, Grammatical Trigrams: A Probabilistic Model of Link Grammar, Presented to the 1992 AAAI Fall Symposium on Probabilistic Approaches to Natural Language. [33] D. Wible, C.-H. Kuo, F.-y. Chien, N.L. Taso, Automating repeated exposure to target vocabulary for second language learners, in: Proceedings of the IEEE International Conference on Advanced Learning Technologies, 6–8 August 2001, pp. 127–128. [34] N. Ide, R. Reppen, K. Suderman, The American National Corpus: more than web can provide, in: Proceedings of the Third Language Resources and Evaluation Conference (LREC), Las Palmas, Canary Islands, Spain, pp. 839–844. [35] N. Ide, The American National Corpus: everything you always wanted to knowldotsand weren’t afraid to ask, Invited keynote, Corpus Linguistics 2003, Lancaster, UK. [36] American National Corpus. Available from . [37] British National Corpus. Available from . [38] Y. Ravin, Grammar errors and style weaknesses in a text-critiquing system, IEEE Trans. Profession. Commun. 31 (3) (1988) 108–115. [39] Microsoft MSDN on line Library. Available from .